BE/BAT 485/585 - Lab-6
Ex-1
Danielle Tadych
Ex-1
Danielle Tadych
In this exercise you will learn and understand
- How hyperspectral differs from Multispectral
- How to Perform Spectral and Spatial Convolution using a sensor RSR information
- How to simulate multispectral data from hyperspectral data
- How to create data for ground truthing
- And finally you will learn how to use the HDF library to open HDF(5) files. Most remote sensing is stored in this format
Convolution is a form of simulating multispectral data from hyperspectral data via the Relative Spectral Responses (RSR). RSRs define the shape of the multispectral sensors and resulting data.
The figure above defines how much each portion of the spectrum contributes to the coarse spectral band.
With this info we can use high definition hyperspectral data to generate or simulate any sensor's spectral band
For example to generate the NIR band (above) we integrate all the hyperspectral data in the range defined by the RSR of the NIR band as would the actual multispectral sensor do.
The spectral convolution process uses the following equation:
$ L_b= \frac { { \int \limits _{\lambda_1} ^{\lambda_2}} L_h.SRF.d { \lambda} } { { \int \limits _{\lambda_1} ^{\lambda_2}} SRF.d { \lambda} } $
Where:
- $\lambda_1$ & $\lambda_2$ are the band's spectral range
- SRF is the RSR value for the portion of the spectrum
- d$\lambda$ is a hyperspectral step
- L$_b$ is the multispectral radiance/reflectance computed
- L$_h$ is the radiance/reflectance for the considered range
The spatial convolution process uses the following equation:
$ V = \frac { 1}{ n.m } \sum \limits _{i=1} ^{n} \sum \limits _{j=1} ^{m} v_{ij} $
Where:
- V is the new pixel (usually coarser) brightness value
- v$_i$$_j$ is the finer resoltuion pixel brightness value
- n and m define the matrix size in the finer resolution image that would correspond to a single pixel in the coarser resoltuion image
The spatial convolution is an aggregation process of the finer resolution pixles to the coarser ones. For example to create a 250m resolution pixel from 1 m pixels we simply need to average 250x250 pixels window. In practice there are other ways to doing this, that we will cover later.
</center>
